Messages in this thread | | | Date | Wed, 20 Feb 2013 17:07:15 +0100 | Subject | Re: [PATCH 2/8] cputime: Librarize per nsecs resolution cputime definitions | From | Frederic Weisbecker <> |
| |
2013/2/14 Kevin Hilman <khilman@linaro.org>: > Frederic Weisbecker <fweisbec@gmail.com> writes: > >> The full dynticks cputime accounting that we'll soon introduce >> will rely on sched_clock(). And its clock can have a per >> nanosecond granularity. >> >> To prepare for this, we need to have a cputime_t implementation >> that has this precision. >> >> ia64 virtual cputime accounting already uses that granularity >> so all we need is to librarize its implementation in the asm >> generic headers. > > ...except that for it to be truly generic (i.e. usable on 32-bit arches), > it should be using do_div() for 64-bit divides. > > I understand this series is targetted at 64-bit platforms currently, but > I'm looking (again) at getting adaptive NOHZ working on ARM, and this > was the first obvious obstacle to even compiling. > > I see this is already merged into tip/sched/core, so below is a patch on > top of that to convert to using do_div(). If you're OK with the > approach, I'll post it formally to LKML.
Yes please, this looks like something we want.
Thanks.
| |